Manit Kaushik

AI Researcher

New Delhi, Delhi, India9 mos experience

AI EnabledAI ML Practitioner

Key Highlights

Developed high-performance AI solutions with TensorRT.
Created multi-cloud interoperability for seamless database operations.
Built a document-based conversational AI chatbot.

Stackforce AI infers this person is a skilled AI and Cloud Computing developer with a focus on machine learning applications.

Contact

Skills

Core Skills

Ai ResearchMachine LearningC++Cloud ComputingSoftware DevelopmentConversational Ai

Other Skills

.NET FrameworkAPIAWSAzure Cosmos DBBLASC#CUDAData AnalysisDatabase Management System (DBMS)FlaskGame DevelopmentInformation RetrievalJUnitJavaLangChain

About

I am a highly motivated Computer Science & Engineering student at IIIT Delhi with a strong academic background and a passion for software development and AI research.

Experience

Research commons

AI Researcher Intern

Aug 2025 – Nov 2025 · 3 mos · Remote

Built cppgrad-tensor, a C++ tensor library with core tensor ops, where native kernels are comparable to BLAS on small inputs and BLAS acceleration delivers up to 160× faster matmul and 230× faster tensordot on large tensors.
Optimized DeepSeek-v1 LLM inference performance using TensorRT LLM, achieving a 49.2% reduction in model’s size while achieving a 5x increase in inference speed (116 tokens/sec vs. 20 tokens/sec) on CNN summarization task, incurring only a 4.5% decrease in ROUGE Lsum Score
Developed a course-aligned RAG agent for a multi-agent buy-side financial analyst, enabling actionable insight extraction from financial resources via query-to-slide matching and voice-over search.

C++TensorRTRetrieval-Augmented Generation (RAG)AI ResearchMachine Learning

Microsoft

Software Engineer Intern

May 2025 – Jul 2025 · 2 mos · Bengaluru, Karnataka, India · On-site

Received PPO
Built a proof-of-concept multi-cloud interoperability layer enabling AWS DynamoDB SQL APIs (CRUD + Query) to run seamlessly on Azure Cosmos DB, requiring no client-side code changes.
Achieved P99 latency within 3.9% and average latency within 9.3% of native DynamoDB by designing a low-latency request translation pipeline with signature-based authorization, despite an additional network hop.

AWSAzure Cosmos DBSQLCloud ComputingSoftware Development

University of glasgow

Undergraduate Researcher

Jun 2024 – Apr 2025 · 10 mos · Remote

Worked under Prof. Debasis Ganguly on a research project titled "Exploring the Role of Diversity in Example Selection for In-Context Learning", which has been accepted in the Short Papers track at ACM SIGIR 2025 (A*) Conference.

Speridian technologies

AI/ML Intern

Feb 2024 – Mar 2024 · 1 mo · Remote

Built and deployed the DocuTalk Chatbot, a document-based conversational AI chatbot leveraging LangChain, a fine-tuned LLaMA2, and PyTesseract OCR, complete with a Streamlit frontend and Flask APIs, validated through Postman.

LangChainFlaskStreamlitConversational AIMachine Learning